| Summary: | 碩士 === 樹德科技大學 === 資訊工程學系 === 94 === The quality issue of VoIP service is significantly attractive, since VoIP has become a promising service on Internet. In this thesis, the effectiveness and performance of random early detection (RED) applied on ADSL routers to control VoIP service quality are investigated. In RED, the drop probability, maximum and minimum thresholds are three operational parameters to determine when and how to drop the packets in the queue of routers. The proper packet dropping will be helpful to reduce the queuing delay and jitter of VoIP packets caused by over-flow and high variation of packet queues. Additionally, the determination of operational parameters depends on the requirement of quality-of-service (QoS), for example differentiated delay and delay jitter. Thus how to determine the parameters is investigated in this thesis. A back-propagation neural network is proposed to learn the specifying the RED parameters. The inputs of this neural network are delay tolerance and available network bandwidth. and the output is the RED parameter. Simulation results demonstrate, even in low available bandwidth and asymmetrical bandwidth situations, the proposed approach effectively improve the delay and jitter of VoIP service packets. This work benefits network managers to enhance VoIP service by properly deploying RED queueing mechanism.
|
|---|
